Technology Review reports today about a new energy aware internet routing process. Some folks at MIT, Carnegie Mellon and Akamai looked into energy price fluctuations and data center loads across the country to see if there could be any cost savings to rerouting data from high cost data centers to lower cost ones. Their results have shown that large Internet companies like Google, Microsoft and Amazon could save up to 40% on their electricity bills – millions of dollars in savings.
They don’t mention any details as to how the ‘smart routing algorithim’ works but my guess is that they are using decision tree learning with a heuristic algorithm like ID3 or C4.5. Their model would probably need some way to factor in arbitrage opportunities within the electricity markets on a real time basis while simultaneously tracking data-center loads. One of the problems they mention that holds this technology back is that most data-centers today lack the ability to throttle power usage with loads – in other words, the servers need to be able to consume a fraction of their full load power when idle.
What are some potential implications for the future?
New market opportunities for those firms that can offer the hardware solutions to make servers ‘energy elastic’.
This could be a boon for the software cloud, as network computing could take advantage of these routing systems to find the least costly resource. Remember back in college when you had to use one of the computer labs? It was basically a room full of computer stations made available for everything from creating presentations to developing software applications. Now imagine if each of the computers in that room costs a different amount to use. The ideal thing to do would be to walk around the room to find the cheapest one – of course making sure that it has the needed capabilities. I can imagine a situation where cloud software would locate the least costly computing resource for a given task anywhere around the country or world just as I would walk around that computer lab searching for the cheapest available station.
I’m speculating here but could this lead to a normalization of electricity prices across the country as data centers compete for locating and using lower cost centers? I was only able to find one 2005 article mentioning that data centers consumer about 1.5% of all U.S. electricity. I think it would be a fair guess that today’s number might be closer to 2% or higher. If so then how would that impact states that currently have lower electricity rates? Lots of potential externalities here to contemplate.
The article also mentions that energy companies could negotiate with large internet firms in advance of expected peak loads to mitigate potential outages or problems. Energy companies could essentially ‘shape’ their loads to prevent failures or outages.
A problem I see with this technology is that if it is used extensively in a decentralized manner then most of the cost savings could evaporate quickly. I know what you are thinking here, “but that is to be expected. Its simple supply and demand here – increase the use of product x over product y until the marginal cost reaches marginal benefit”.
Here is my problem:
If every party acted independently and outside of a central market or clearinghouse then the anticipated savings would be lost to transactions costs. These transactions costs being those associated with moving data from one place to another and doing so with incomplete information. Take for example if an underutilized data center pops up on the radar of ten or so Internet companies. They each, independently and unbeknownst to the others, make the move to the new data center. This surge in load may lead to an increase in costs and therefore mute any cost savings. There could even be a potential for some Internet companies to have to move yet again - further increasing the cost of the move.
I think that a centralized market or clearinghouse for this type of transaction would facilitate this form of intelligent routing. Of course, I am no expert in data center economics or how or if some of these problems or opportunities are realistic. But with my limited knowledge they would be questions I would ask.